public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
From: "burnus at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug middle-end/100059] New: [OpenMP] wrong code with 'declare target link' and a scalar variable
Date: Tue, 13 Apr 2021 08:58:51 +0000	[thread overview]
Message-ID: <bug-100059-4@http.gcc.gnu.org/bugzilla/> (raw)

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100059

            Bug ID: 100059
           Summary: [OpenMP] wrong code with 'declare target link' and a
                    scalar variable
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Keywords: openmp, wrong-code
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: burnus at gcc dot gnu.org
                CC: jakub at gcc dot gnu.org
  Target Milestone: ---
            Target: nvptx-none

Created attachment 50578
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50578&action=edit
Testcase, compile with 'gcc -fopenmp' with an (nvptx) offloading enabled
compiler

The attached testcase has four global vars:
  int a[N], b[N], c[N];
  int i = 0;

which are used with:

  #pragma omp declare target link(a,c,b,i)  

and

  #pragma omp target map(to: i) map(tofrom: a, b, c) 
    update();

After the call, the one array is not updated but has the previous host value:
* For 'declare target link(a,c,b,i)': not updated array is 'b'
* For 'declare target link(a,b,c,i)': not updated array is 'c'

Namely, the testcase shows:
i=5: A=5, B=6, C=7    // Original host value
i=5: A=6, B=6, C=10   // target call to 'update': 'B' is not updated
i=5: A=7, B=8, C=13   // host update, OK


Additionally, when using 'for (int i' in 'update': The error only shows up if
there is 'i'
And it occurs for the item before 'i', namely:
  link(a,c,b,i) → 'b' is wrong
  link(a,c,i,b) → 'c' is wrong


The issue occurs with nvptx. It works without offloading and I think it works
with GCN (did not show up in the test logs, but I have not manually verified).

             reply	other threads:[~2021-04-13  8:58 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-04-13  8:58 burnus at gcc dot gnu.org [this message]
2021-04-13 13:16 ` [Bug middle-end/100059] " burnus at gcc dot gnu.org
2021-04-13 13:37 ` burnus at gcc dot gnu.org
2021-04-13 14:11 ` burnus at gcc dot gnu.org
2021-04-13 16:39 ` burnus at gcc dot gnu.org
2022-03-29 10:23 ` burnus at gcc dot gnu.org
2023-09-04  9:37 ` tschwinge at gcc dot gnu.org
2023-09-04 12:57 ` [Bug other/100059] " tschwinge at gcc dot gnu.org

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=bug-100059-4@http.gcc.gnu.org/bugzilla/ \
    --to=gcc-bugzilla@gcc.gnu.org \
    --cc=gcc-bugs@gcc.gnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).